Man-machine anti-collision early warning identification method suitable for underground mine trackless equipment

Abstract

The invention discloses a man-machine anti-collision early warning recognition algorithm suitable for underground mine trackless equipment, and belongs to the technical field of mine safety. According to the algorithm, through cooperative sensing of a millimeter wave radar and a UWB sensor, in combination with distance hierarchical identification, a triple verification mechanism of radar ranging, visual identification and infrared temperature is constructed, dynamic weight data fusion and hierarchical early warning intervention are performed, and accurate identification and dynamic risk prevention and control of personnel targets in a complex environment are realized. The anti-collision system has the self-learning capability, can adapt to severe mine environments, and remarkably improves the reliability and adaptability of the anti-collision system.

Description

Technical Field

[0001] This invention relates to the field of underground mine safety engineering technology, specifically to a human-machine collision avoidance early warning and identification method applicable to trackless equipment in underground mines. Background Technology

[0002] Current collision avoidance technologies for human-machine interfaces in underground mines mainly focus on the core logic of "positioning + ranging + early warning," forming a technical system based on single-sensor positioning, simple data overlay, and fixed threshold early warning. Some solutions attempt to introduce multi-sensor fusion or AI-assisted recognition, but a mature closed loop adapted to the complex mining environment has not yet been formed. The mainstream technical paths can be divided into three categories: first, positioning-based early warning technology centered on UWB / RFID; second, ranging-based collision avoidance technology centered on radar / laser; and third, a preliminary composite solution integrating positioning and ranging, but with a low level of intelligence. These three technical approaches have the following problems: 1. Poor coordination between positioning and ranging, with shortcomings in full-range coverage: UWB solutions have insufficient dynamic response speed within 50 meters (detection frequency only 10Hz), failing to capture the rapid movement of personnel / equipment; radar solutions show significant signal attenuation above 50 meters, resulting in failure at medium and long ranges; preliminary composite solutions only perform data superposition without dynamically allocating sensor weights according to distance ranges, leading to lag in medium and long-range positioning and susceptibility to interference in short-range ranging; 2. Low target recognition accuracy, with prominent false alarms and missed alarms: UWB / RFID solutions can only locate coordinates and cannot distinguish between personnel and static obstacles such as rocks and equipment, easily triggering false alarms; single radar solutions rely on RCS values ​​or distance judgments, making it difficult to avoid interference from metal structures and abandoned equipment in tunnels; some AI-assisted solutions can improve the recognition rate, but require a large amount of data training and lack robustness in high-dust and completely dark environments; 3. Rigid early warning algorithms, unable to adapt to complex scenarios: alarms are triggered only based on distance or a single speed parameter, without considering equipment type (high-speed transport / Differences in low-speed loading and tunnel environment (curves / slope changes); for example, mining trucks and loaders use the same warning threshold, resulting in untimely warnings from high-speed equipment and frequent false alarms from low-speed equipment. Although some solutions support parameter adjustment, they require manual configuration, resulting in extremely poor adaptability; 4. Lack of self-learning and adaptive capabilities, and insufficient long-term stability: When faced with environmental changes such as changes in dust concentration and tunnel structure modifications, parameters need to be manually recalibrated, otherwise the recognition accuracy will drop significantly; and the algorithm cannot be optimized through historical data, so the false alarm rate gradually increases with environmental changes during long-term operation; 5. Disconnect between warning and intervention, and insufficient safety protection levels: Most solutions remain at the level of "audio-visual alarms" and lack effective equipment intervention mechanisms: even if a warning is triggered, it still relies on manual operation by the driver to avoid risks, which can easily lead to accidents in scenarios where the driver is fatigued or has limited visibility; a few solutions support equipment braking, but the triggering conditions are singular (based only on a fixed distance), which can easily lead to over-intervention or untimely intervention. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention provides a human-machine collision avoidance early warning and identification method applicable to trackless equipment in underground mines.

[0004] The specific solution of the present invention is as follows: A human-machine collision avoidance early warning and identification method suitable for trackless equipment in underground mines, comprising the following steps: A human-machine collision avoidance early warning and identification method suitable for trackless equipment in underground mines, comprising the following steps:

[0005] S1. Employs millimeter-wave radar and UWB sensors for collaborative sensing, dynamically switching the dominant sensor based on distance range;

[0006] S2. Target recognition based on distance hierarchy, including long-range UWB positioning, mid-range dual-sensor fusion, and close-range triple verification;

[0007] S3. Data fusion is performed using a hierarchical fusion and dynamic weighting mechanism;

[0008] S4. Based on the two-dimensional parameters of "distance-speed", a collision risk coefficient is constructed to realize three-level early warning and graded intervention;

[0009] S5. Optimize model parameters using historical data.

[0010] Furthermore, in step S2, target identification based on distance stratification specifically includes: long-range identification within the range of 50-100 meters, medium-range identification within the range of 10-50 meters, and short-range identification within the range of 0-10 meters; long-range identification uses UWB positioning data as the core, and calculates the relative position of the UWB tag on the personnel end and the UWB tag on the equipment end through the time difference of arrival algorithm; medium-range identification adopts dual-sensor data fusion, with UWB providing high-frequency updated position coordinates, and millimeter-wave radar capturing the real-time distance and relative velocity of the target through fan-shaped scanning to form a two-dimensional data pair of "position-velocity"; short-range identification uses millimeter-wave radar data as the core, increases the detection frequency to 30Hz, and simultaneously links high-definition cameras and infrared sensors for target verification.

[0011] Furthermore, in step S2, the triple verification includes a combination of three identification methods: radar ranging, visual recognition, and infrared temperature measurement, to ensure that the accuracy of personnel target identification is ≥99%.

[0012] Furthermore, in step S3, the dynamic weights are allocated according to the distance range: for distances above 50 meters, the weight for UWB is 0.8 and the weight for millimeter-wave radar is 0.2; for distances between 10 and 50 meters, the weights for both UWB and millimeter-wave radar are 0.5; and for distances within 10 meters, the weight for millimeter-wave radar is 0.8 and the weight for UWB is 0.2.

[0013] Furthermore, in step S4, the formula for calculating the collision risk factor R is: In the formula, V is the relative speed between equipment and personnel, D is the real-time distance, D1 is the warning start threshold of the current distance interval, α is the relative speed weight, β is the distance weight, and α and β are dynamically adjusted according to the equipment type and the roadway environment.

[0014] Furthermore, the warning start threshold is divided into three levels: Level 1 warning start threshold D1=50m, Level 2 warning start threshold D1=20m, and Level 3 warning start threshold D1=10m.

[0015] Furthermore, the equipment is divided into high-speed equipment with a speed greater than 15 km / h and low-speed equipment with a speed less than or equal to 15 km / h. For high-speed equipment, α=0.7 and β=0.3; for low-speed equipment, α=0.5 and β=0.5. The tunnel environment includes tunnel curvature; if the curvature is greater than 50 m... -1 If so, then β will increase by 0.1.

[0016] Furthermore, the three-level early warning system includes a level one safety alert, a level two proactive warning, and a level three mandatory intervention. A level one safety alert is triggered when D ≥ 20m and R < 0.4, a level two proactive warning is triggered when 10m ≤ D < 20m and 0.4 ≤ R < 0.7, and a level three mandatory intervention is triggered when D < 10m and R ≥ 0.7.

[0017] Furthermore, step S5 specifically involves: automatically recording the environmental parameters, equipment operating status, and target detection data at the time of each warning event; optimizing the weight coefficients α, β, and threshold parameters in the warning model using a gradient descent algorithm; and periodically updating the model based on historical data to ensure adaptability in different mining environments.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0019] Compared with existing technologies, this invention has the following advantages: 1. It achieves accurate coverage across the entire distance (0-100 meters), solving the problem of poor coordination between positioning and ranging; 2. Through a triple verification mechanism, it significantly improves the accuracy of personnel target identification and reduces false alarms and missed alarms; 3. The early warning algorithm dynamically adapts to equipment type and tunnel environment, avoiding the defects of fixed thresholds; 4. It has self-learning ability, adapts to complex environmental changes, and has high long-term stability; 5. It constructs a "early warning-intervention" closed loop to achieve proactive safety protection. Attached Figure Description

[0020] Figure 1 is a schematic diagram of the overall architecture of the present invention;

[0021] Figure 2 is a flowchart of the present invention. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1 , Figure 2 The specific embodiments of the present invention will be further described below.

[0023] like Figure 1 As shown, this embodiment uses an underground mining loader as an example to illustrate the specific implementation of the present invention. The system hardware includes:

[0024] Vehicle-mounted warning terminal: Installed in the driver's cab, it integrates a data processing unit, a warning logic module, a human-machine interface (touch screen), and a CAN bus communication interface.

[0025] Collision avoidance detection terminals: A total of 4 sets are installed, one each on the front, rear, left, and right sides of the loader. Each terminal includes:

[0026] Millimeter-wave radar (77GHz, detection angle 120°, maximum detection range 80 meters, ranging accuracy ±0.1 meters, data update frequency 20Hz); UWB receiver module (operating frequency 3.5~6.5GHz, positioning accuracy ±0.3 meters, update frequency 10Hz); infrared thermal imaging sensor (temperature measurement range 0~60℃, accuracy ±0.5℃); high-definition camera (2 megapixels, with infrared supplementary light); audible and visual alarm device: including a three-color warning light (red / yellow / green), a buzzer, and a driver's seat vibration motor, installed inside and outside the driver's cab.

[0027] Personnel-worn equipment: UWB tag (worn on a helmet or belt), with a built-in vibration motor and a small audible and visual alarm.

[0028] Communication network: All on-board devices are interconnected via CAN bus (ISO 11898); UWB tags and on-board UWB modules use wireless pulse communication.

[0029] like Figure 2 As shown, this embodiment provides the above-mentioned human-machine collision avoidance warning and recognition method for loader trucks, which specifically includes the following steps:

[0030] S1. System initialization and environment matching:

[0031] The vehicle-mounted terminal loads a preset electronic map of the alleyway (including alleyway boundaries, slope, and curvature information).

[0032] The UWB base station (with key locations in the alleyway pre-set) and the vehicle-mounted UWB module complete time synchronization.

[0033] The system automatically initializes the warning parameters based on the current vehicle type (mining truck): Level 1 warning starting distance D1=50m, Level 2 warning starting distance D1=20m, Level 3 warning starting distance D1=10m, speed weight α=0.7, and distance weight β=0.3.

[0034] S2. Layered target identification and data acquisition:

[0035] (1) Long-distance identification (50-100 meters): Based on UWB positioning data, the relative position of the UWB tag at the personnel end and the UWB tag at the equipment end is calculated through the TDOA (Time Difference of Arrival) algorithm. The system presets an electronic map of the tunnel, matches the UWB positioning coordinates with the map, and removes invalid data outside the tunnel boundary; at the same time, it judges the personnel status based on the location data of 5 consecutive seconds. If the position fluctuation range is <0.5 meters, it is marked as a "static target" and the warning priority is reduced.

[0036] (2) Mid-range identification (10-50 meters): Initiate dual-sensor data fusion. UWB provides frequently updated position coordinates, and millimeter-wave radar captures the target's real-time distance and relative speed through fan-shaped scanning (detection angle 120°), forming a "position-velocity" two-dimensional data pair. The system fuses the two sets of data using a Kalman filter algorithm to correct the slight lag error of UWB positioning and the random interference of millimeter-wave radar, outputting a smooth target trajectory; if the angle between the trajectory and the equipment's driving trajectory is <30°, it is determined to be a "potential collision target".

[0037] (3) Close-range identification (0-10 meters): Based on millimeter-wave radar data, the detection frequency is increased to 30Hz, and a high-definition camera and infrared sensor are linked to verify the target. The millimeter-wave radar outputs the target's distance, relative velocity, and radar cross-section (RCS). The RCS value of personnel targets is usually between -10 and 5dBsm. Based on this, the system eliminates interfering targets such as equipment and rocks with an RCS value >10dBsm. The camera identifies the target outline through the HOG human feature algorithm and combines it with the temperature data of the infrared sensor (human body temperature 35-38℃) to form a triple confirmation mechanism of "radar ranging + visual recognition + temperature verification" to ensure that the accuracy of personnel target identification is ≥99%.

[0038] S3, Data Fusion:

[0039] By deeply fusing dual-source data from UWB and millimeter-wave radar, the advantages of both in position stability and dynamic response speed are integrated, the detection error of a single sensor is corrected, and accurate target position, relative velocity, and trajectory data are output, providing reliable data support for collision risk assessment.

[0040] The system adopts a "layered fusion + dynamic weighting" architecture: the bottom layer is sensor data preprocessing, which removes outliers (such as millimeter-wave radar single ranging error > 1 meter, UWB positioning jump > 1 meter) through threshold filtering; the middle layer is feature layer fusion, which transforms UWB position data and millimeter-wave radar distance and velocity data into a unified three-dimensional spatial data format; the top layer is decision layer fusion, which dynamically allocates weights based on distance range—UWB weight is 0.8 and millimeter-wave radar weight is 0.2 for distances above 50 meters (emphasizing position stability); UWB and millimeter-wave radar weights are each 0.5 for distances between 10 and 50 meters (balancing position and velocity accuracy); and millimeter-wave radar weight is 0.8 and UWB weight is 0.2 for distances below 10 meters (emphasizing dynamic response speed). The final output is the fused target position, relative velocity, and trajectory data.

[0041] S4. Collision Risk Calculation and Warning Triggering:

[0042] A warning algorithm model is constructed based on the two-dimensional core parameters of "distance-speed". A three-level warning threshold is set by combining the equipment driving status and the characteristics of the roadway environment to achieve precise matching between warning intensity and collision risk.

[0043] (1) Core Early Warning Algorithm Model

[0044] The optimized model for calculating the collision risk factor R is as follows: The definitions and values ​​of each parameter are as follows:

[0045] a) Basic parameters: V is the relative speed between equipment and personnel (m / s, directly detected by millimeter-wave radar or calculated by UWB position change rate), D is the real-time distance (m, fused UWB and millimeter-wave radar data), and D1 is the warning start threshold for the current distance range.

[0046] b) Dynamic weights: α (relative speed weight) and β (distance weight) are dynamically adjusted according to equipment type and roadway environment—high-speed equipment such as mining trucks (>15km / h): α=0.7, β=0.3; low-speed equipment such as loaders (≤15km / h): α=0.5, β=0.5; curved roadways (curvature >50m) -1 The β weight was increased by 0.1 to ensure the dominant role of the distance factor. The weight was determined by optimizing a neural network model trained on historical accident data.

[0047] (2) Tiered warning thresholds and triggering logic

[0048] a) Level 1 Warning (Safety Alert): Triggered when D ≥ 20m and R < 0.4. Upon triggering, the green warning light on the equipment remains constantly lit, and the vehicle screen displays the personnel's position and distance; the personnel's wristband vibrates at a low frequency of 2Hz without emitting any audible or visual alarms to avoid interfering with normal operations. The warning automatically cancels when D increases to 50 meters or more or R < 0.2.

[0049] b) Level 2 Warning (Active Warning): Triggering conditions are 10m ≤ D < 20m and 0.4 ≤ R < 0.7. The yellow warning light on the equipment flashes at a frequency of 5Hz, the buzzer emits a "beep beep" sound at a frequency of 1Hz, and the driver's seat vibrates slightly at a frequency of 3Hz; the personnel's wristband vibrates at a high frequency of 5Hz, and the audible and visual alarm plays a loop of the voice prompt "Equipment approaching, please give way." If D does not increase or R does not decrease within 10 seconds, the warning automatically escalates to Level 3.

[0050] c) Level 3 Warning (Forced Intervention): Triggering conditions are D < 10m and R ≥ 0.7. The red warning light on the equipment side flashes rapidly at a frequency of 10Hz, the buzzer sounds continuously, and the driver's seat vibrates strongly; the audible and visual alarm on the personnel side continuously sounds at a high decibel of 110dB. Simultaneously, the equipment control unit intervenes. If D ≥ 5m and R < 0.9, the equipment is controlled to decelerate to below 5km / h; if D < 5m or R ≥ 0.9, and there is no risk mitigation within 1 second, forced braking is triggered until the equipment comes to a complete stop.

[0051] S5. Optimize model parameters using historical data:

[0052] The system possesses self-learning capabilities. After each warning event (including false alarms and missed alarms), it automatically records the environmental parameters (dust concentration, tunnel curvature, light intensity), equipment operating status (speed, turning), and target detection data. It then optimizes the weight coefficients α, β, and threshold parameters in the warning model using a gradient descent algorithm. The model is updated monthly based on historical data to ensure its adaptability to different mining environments, with a false alarm rate below 1% and a missed alarm rate of 0.

[0053] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

Claims

1. A human-machine collision avoidance early warning and identification method applicable to trackless equipment in underground mines, characterized by: Includes the following steps: S1. Employs millimeter-wave radar and UWB sensors for collaborative sensing, dynamically switching the dominant sensor based on distance range; S2. Target recognition based on distance hierarchy, including long-range UWB positioning, mid-range dual-sensor fusion, and close-range triple verification; S3. Data fusion is performed using a hierarchical fusion and dynamic weighting mechanism; S4. Based on the two-dimensional parameters of "distance-speed", a collision risk coefficient is constructed to realize three-level early warning and graded intervention; S5. Optimize model parameters using historical data.

2. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 1, characterized in that: In step S2, target identification based on distance stratification specifically includes: long-range identification within the range of 50-100 meters, medium-range identification within the range of 10-50 meters, and short-range identification within the range of 0-10 meters. Long-range identification uses UWB positioning data as the core and calculates the relative position of the UWB tag on the personnel end and the UWB tag on the equipment end through the time difference of arrival algorithm. Medium-range identification adopts dual-sensor data fusion. UWB provides high-frequency updated position coordinates, and millimeter-wave radar captures the real-time distance and relative velocity of the target through fan-shaped scanning to form a two-dimensional data pair of "position-velocity". Short-range identification uses millimeter-wave radar data as the core, increases the detection frequency to 30Hz, and simultaneously links high-definition cameras and infrared sensors for target verification.

3. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 1, characterized in that: In step S2, the triple verification includes a combination of three identification methods: radar ranging, visual recognition, and infrared temperature measurement, to ensure that the accuracy of personnel target identification is ≥99%.

4. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 1, characterized in that: In step S3, the dynamic weights are allocated according to the distance range: for distances above 50 meters, the weight for UWB is 0.8 and the weight for millimeter-wave radar is 0.2; for distances between 10 and 50 meters, the weights for both UWB and millimeter-wave radar are 0.5; and for distances within 10 meters, the weight for millimeter-wave radar is 0.8 and the weight for UWB is 0.

2.

5. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 1, characterized in that: In step S4, the formula for calculating the collision risk factor R is: In the formula, V is the relative speed between equipment and personnel, D is the real-time distance, D1 is the warning start threshold of the current distance interval, α is the relative speed weight, β is the distance weight, and α and β are dynamically adjusted according to the equipment type and the roadway environment.

6. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 5, characterized in that: The warning start threshold is divided into three levels: Level 1 warning start threshold D1=50m, Level 2 warning start threshold D1=20m, and Level 3 warning start threshold D1=10m.

7. The human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 5, characterized in that: The equipment is divided into high-speed equipment with a speed greater than 15 km / h and low-speed equipment with a speed less than or equal to 15 km / h. For high-speed equipment, α=0.7 and β=0.3; for low-speed equipment, α=0.5 and β=0.

5. The tunnel environment includes tunnel curvature. If the curvature is greater than 50 m... -1 If so, then β will increase by 0.

1.

8. A human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 5, characterized in that: The three-level early warning system includes a level one safety alert, a level two proactive warning, and a level three mandatory intervention. A level one safety alert is triggered when D ≥ 20m and R < 0.4; a level two proactive warning is triggered when 10m ≤ D < 20m and 0.4 ≤ R < 0.7; and a level three mandatory intervention is triggered when D < 10m and R ≥ 0.

7.

9. A human-machine collision avoidance early warning and identification method for trackless equipment in underground mines according to claim 1, characterized in that: Step S5 specifically involves: automatically recording the environmental parameters, equipment operating status, and target detection data at the time of each warning event, optimizing the weight coefficients α, β, and threshold parameters in the warning model through the gradient descent algorithm, and periodically updating the model based on historical data to ensure adaptability in different mining environments.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by the processor, the program implements the human-machine collision avoidance warning and identification method for trackless equipment in underground mines as described in any one of claims 1-9.

Images